Agentic AI, LLM engineering, and the models behind modern automation — multi-agent systems, LLM evaluation and comparisons, RAG, fine-tuning, AI infrastructure, security, and production AI engineering.
From the blog
Neon's branch-per-PR model is built for agent workloads — 80% of Neon DBs are now provisioned by AI agents. Spin up a database in 500ms, run an experiment, throw it away. Working Drizzle + Neon code.
Conversation Orchestrator + Memory + Intelligence (GA at SIGNAL 2026) close the channel-switch gap. We design CallSphere's omnichannel pattern: voice + SMS + WhatsApp on one Conversation, with persistent memory.
Traditional RPS rate limits fail against LLM-driven voice. A single 30s call can burn 8K tokens. Here is the 2026 token-aware rate-limit pattern that keeps cost predictable across 50K concurrent calls.
Most use cases that 'need fine-tuning' actually need a better prompt. We give you a 90-second decision tree across data availability, taxonomy churn, latency, and total-cost-per-correct-decision — backed by IBM's 2026 framework and CallSphere's real production calls.
Safari 26.4 pushed WebTransport into Baseline status. HTTP/3 + QUIC kills head-of-line blocking and matches WebRTC datagram latency without the SDP. Where it fits in voice AI architecture.
Reduce international VoIP call latency for distributed teams. Codec selection, geographic routing, TURN placement, and carrier optimization strategies.
How to authenticate a WebSocket on upgrade, reconnect cleanly with exponential backoff and jitter, and refresh long-lived tokens without dropping the session.
How AI voice agents earn A-level STIR/SHAKEN attestation, what the 2026 FCC robocall mitigation rules require, and the operational checklist for AI callers.
WebRTC dominates browser AI voice for a reason - UDP under the hood. But SIP over WebSocket still wins for click-to-call inside a SaaS app where you control both ends. Here is the 2026 picture.
